National Institute of Technology Nagaland

Human gut microbiomes respond to lifestyle transitions, yet the extent to which these responses are conserved across spatio-cultural contexts remains undercharacterized. We present the South Asian MicroBiome ARray (SAMBAR), a population-scale 16S gut microbiome study of 575 adults from ten geographically and socio-culturally diverse South Asian communities. Each community was sampled in ancestral villages and urban centers, enabling controlled comparisons of geography and lifestyle. Relative to global cohorts, SAMBAR microbiomes occupy a distinct compositional space with stronger correlation to geography and community membership than lifestyle. Although urbanization is consistently associated with increased abundance of disease-linked taxa, microbiome responses to lifestyle transitions are largely community-driven, including the acquisition of wheat- and dairying-associated microbial modules in some communities that may facilitate non-genetic adaptation to lactase non-persistence. Microbiome responses to urbanization are heterogeneous even at regional scales, reflecting local culture and geography and underscoring the need for community-specific investigations of health impacts.

Landslides pose severe risks to human life, infrastructure, and property, particularly in regions with heavy rainfall and unstable soil. Existing monitoring methods often fail to provide real-time detection and early warnings, increasing the potential for catastrophic damages. This work presents edge analytics using IoT devices as a solution for automated landslide detection and early warning systems in a cost-effective and scalable approach. The system integrates an MPU6050 accelerometer-gyroscope for detecting soil movement, a soil moisture sensor for monitoring saturation levels, and a Raspberry Pi camera for image-based terrain analysis using OpenCV. By analyzing sensor data and detecting soil cracks through image processing, the system triggers alerts when predefined risk thresholds are exceeded. Controlled experiments validate the system's reliability in identifying early warning signs of landslides. The proposed system offers a practical and efficient solution for continuous landslide monitoring, with potential applications in disaster preparedness and risk mitigation for landslide-prone areas.

This study presents a novel deep learning approach addressing the critical shortage of veterinary expertise in India’s North Eastern Hill (NEH) region through automated identification of parasitic infections in livestock. We developed a Convolutional Neural Network (CNN) architecture capable of analyzing both standard and microscopic images to identify and classify 16 distinct parasitic species. The model comprises four convolutional layers (32, 64, 128, 256 filters) with ReLU activation and MaxPooling for efficient feature extraction, followed by Dense layers and a Softmax classifier. The model was trained on a comprehensive dataset of over 5,334 annotated images, achieving 96% accuracy after 30 training epochs. To evaluate stability, it was trained ten times, yielding an average accuracy of 0.9616 ± 0.0024 (95% CI: [0.9601, 0.9630]), Macro F1 of 0.9527 ± 0.0021, and Weighted F1 of 0.9598 ± 0.0019, demonstrating consistent performance. A PHP-based web interface enables real-time predictions and adaptable deployment across hardware and cloud platforms. This system offers a scalable and accessible diagnostic tool for enhancing parasite detection and livestock health monitoring.

ABSTRACT Fascinating features of 2D graphene oxide (GO) have very high potential to enhance the mechanical performance and fracture toughness of epoxy composites after successful dispersion in the epoxy matrix. The dispersion of GO in a highly viscous epoxy matrix was achieved by lateral exfoliation and longitudinal size reduction of graphene sheets. GO was synthesized via an improved Hummer's method, exfoliated through dual‐mode ultrasonication for 20, 30, and 60 min, and reduced to nanoscale (~217 to 227 nm) using high‐energy wet ball milling for 9 h. The synthesized GO and BGO were characterized using XRD, FT‐IR, Raman spectroscopy, FESEM, and TGA to confirm crystal structure, functional groups, defect density, surface morphology, and thermal stability respectively. It was observed that even at higher filler loadings of 1 wt.% of GO and BGO exfoliated for 60 min (GO60@1% and BGO60@1%), the resulting nanocomposites showed a substantial improvement in tensile strength by 39.92% and 41.91%, respectively, and modulus by 6.92% and 12.90%, respectively compared to the NE. Whereas, incorporation of GO60@1% and BGO60@1% into epoxy matrix significantly enhanced fracture toughness ( K IC ) by ~72% and ~140% respectively and the fracture energy ( G IC ) by ~322% and ~356%, respectively with respect to NE. The enhancement is attributed to the superior dispersibility of BGO, which facilitates strong and consistent interfacial interactions with the epoxy matrix, thereby improving the material's resistance to fracture. The exfoliation of GO through ball milling can be a suitable method for industries to prepare graphene‐based nanocomposites with significantly high tensile and fracture properties.

Due to evolution in the field of Artificial Intelligence, learning is more oriented toward videos, more than reading pages of text. Not only in the field of education, entertainment, politics, etc., but in all the domains there are huge number of videos which are uploaded by various users. Hence it is very much necessary to have an efficient video classification system to classify videos into various categories and to make users view their own interesting videos without spending more time searching the same on internet. Proposed framework combines spatial learning strength of VGG16 with temporal learning concept of stacked LSTM layers. Honey Badger Optimization algorithm is used to optimize the model convergence by fine tuning the hyperparameters. An Attention mechanism further improves the framework by assigning greater weights to the frames which extracts more information from subtle motion patterns. Detailed ablation experiments show that introducing ensembled technique on YouTube 8M dataset depicting VIC-FRAME++ consistently surpasses standard CNN-LSTM and Efficient Net techniques in metrics of precision, recall and F1 score. Hence the proposed framework is implemented for real world uses like video organization and recommendation.